Compare commits

...

2 Commits

Author SHA1 Message Date
mathieuRA
d12ece455d feat(xo-web/backup): add 'nRetriesVmBackupFailures' input 2024-02-22 17:42:54 +01:00
mathieuRA
bdc6e43ff2 feat(@xen-orchestra/backups): retry if VM fails 2024-02-22 17:42:54 +01:00
8 changed files with 176 additions and 36 deletions

View File

@@ -6,11 +6,12 @@ import { extractIdsFromSimplePattern } from '../extractIdsFromSimplePattern.mjs'
import { Task } from '../Task.mjs'
import createStreamThrottle from './_createStreamThrottle.mjs'
import { DEFAULT_SETTINGS, Abstract } from './_Abstract.mjs'
import { runTask } from './_runTask.mjs'
import { getAdaptersByRemote } from './_getAdaptersByRemote.mjs'
import { FullRemote } from './_vmRunners/FullRemote.mjs'
import { IncrementalRemote } from './_vmRunners/IncrementalRemote.mjs'
const noop = Function.prototype
const DEFAULT_REMOTE_VM_SETTINGS = {
concurrency: 2,
copyRetention: 0,
@@ -20,6 +21,7 @@ const DEFAULT_REMOTE_VM_SETTINGS = {
healthCheckVmsWithTags: [],
maxExportRate: 0,
maxMergedDeltasPerRun: Infinity,
nRetriesVmBackupFailures: 0,
timeout: 0,
validateVhdStreams: false,
vmTimeout: 0,
@@ -41,6 +43,7 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
const throttleStream = createStreamThrottle(settings.maxExportRate)
const config = this._config
await Disposable.use(
() => this._getAdapter(job.sourceRemote),
() => (settings.healthCheckSr !== undefined ? this._getRecord('SR', settings.healthCheckSr) : undefined),
@@ -62,8 +65,19 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
const allSettings = this._job.settings
const baseSettings = this._baseSettings
const queue = new Set(vmsUuids)
const taskByVmId = {}
const nTriesByVmId = {}
const handleVm = vmUuid => {
if (nTriesByVmId[vmUuid] === undefined) {
nTriesByVmId[vmUuid] = 0
}
nTriesByVmId[vmUuid]++
const taskStart = { name: 'backup VM', data: { type: 'VM', id: vmUuid } }
const vmSettings = { ...settings, ...allSettings[vmUuid] }
const isLastRun = nTriesByVmId[vmUuid] === vmSettings.nRetriesVmBackupFailures + 1
const opts = {
baseSettings,
@@ -72,7 +86,7 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
healthCheckSr,
remoteAdapters,
schedule,
settings: { ...settings, ...allSettings[vmUuid] },
settings: vmSettings,
sourceRemoteAdapter,
throttleStream,
vmUuid,
@@ -86,10 +100,39 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
throw new Error(`Job mode ${job.mode} not implemented for mirror backup`)
}
return runTask(taskStart, () => vmBackup.run())
if (taskByVmId[vmUuid] === undefined) {
taskByVmId[vmUuid] = new Task(taskStart)
}
const task = taskByVmId[vmUuid]
return task
.run(async () => {
try {
const result = await vmBackup.run()
task.success(result)
return result
} catch (error) {
if (isLastRun) {
throw error
} else {
Task.warning(`Retry the VM mirror backup due to an error`, {
attempt: nTriesByVmId[vmUuid],
error: error.message,
})
queue.add(vmUuid)
}
}
})
.catch(noop)
}
const { concurrency } = settings
await asyncMapSettled(vmsUuids, !concurrency ? handleVm : limitConcurrency(concurrency)(handleVm))
const _handleVm = !concurrency ? handleVm : limitConcurrency(concurrency)(handleVm)
while (queue.size > 0) {
const vmIds = Array.from(queue)
queue.clear()
await asyncMapSettled(vmIds, _handleVm)
}
}
)
}

View File

@@ -11,6 +11,8 @@ import { getAdaptersByRemote } from './_getAdaptersByRemote.mjs'
import { IncrementalXapi } from './_vmRunners/IncrementalXapi.mjs'
import { FullXapi } from './_vmRunners/FullXapi.mjs'
const noop = Function.prototype
const DEFAULT_XAPI_VM_SETTINGS = {
bypassVdiChainsCheck: false,
checkpointSnapshot: false,
@@ -24,6 +26,7 @@ const DEFAULT_XAPI_VM_SETTINGS = {
healthCheckVmsWithTags: [],
maxExportRate: 0,
maxMergedDeltasPerRun: Infinity,
nRetriesVmBackupFailures: 0,
offlineBackup: false,
offlineSnapshot: false,
snapshotRetention: 0,
@@ -53,6 +56,7 @@ export const VmsXapi = class VmsXapiBackupRunner extends Abstract {
const throttleStream = createStreamThrottle(settings.maxExportRate)
const config = this._config
await Disposable.use(
Disposable.all(
extractIdsFromSimplePattern(job.srs).map(id =>
@@ -89,48 +93,98 @@ export const VmsXapi = class VmsXapiBackupRunner extends Abstract {
const allSettings = this._job.settings
const baseSettings = this._baseSettings
const queue = new Set(vmIds)
const taskByVmId = {}
const nTriesByVmId = {}
const handleVm = vmUuid => {
const getVmTask = () => {
if (taskByVmId[vmUuid] === undefined) {
taskByVmId[vmUuid] = new Task(taskStart)
}
return taskByVmId[vmUuid]
}
const vmBackupFailed = error => {
if (isLastRun) {
throw error
} else {
Task.warning(`Retry the VM backup due to an error`, {
attempt: nTriesByVmId[vmUuid],
error: error.message,
})
queue.add(vmUuid)
}
}
if (nTriesByVmId[vmUuid] === undefined) {
nTriesByVmId[vmUuid] = 0
}
nTriesByVmId[vmUuid]++
const vmSettings = { ...settings, ...allSettings[vmUuid] }
const taskStart = { name: 'backup VM', data: { type: 'VM', id: vmUuid } }
const isLastRun = nTriesByVmId[vmUuid] === vmSettings.nRetriesVmBackupFailures + 1
return this._getRecord('VM', vmUuid).then(
disposableVm =>
Disposable.use(disposableVm, vm => {
taskStart.data.name_label = vm.name_label
return runTask(taskStart, () => {
const opts = {
baseSettings,
config,
getSnapshotNameLabel,
healthCheckSr,
job,
remoteAdapters,
schedule,
settings: { ...settings, ...allSettings[vm.uuid] },
srs,
throttleStream,
vm,
}
let vmBackup
if (job.mode === 'delta') {
vmBackup = new IncrementalXapi(opts)
} else {
if (job.mode === 'full') {
vmBackup = new FullXapi(opts)
} else {
throw new Error(`Job mode ${job.mode} not implemented`)
Disposable.use(disposableVm, async vm => {
if (taskStart.data.name_label === undefined) {
taskStart.data.name_label = vm.name_label
}
const task = getVmTask()
return task
.run(async () => {
const opts = {
baseSettings,
config,
getSnapshotNameLabel,
healthCheckSr,
job,
remoteAdapters,
schedule,
settings: vmSettings,
srs,
throttleStream,
vm,
}
}
return vmBackup.run()
})
let vmBackup
if (job.mode === 'delta') {
vmBackup = new IncrementalXapi(opts)
} else {
if (job.mode === 'full') {
vmBackup = new FullXapi(opts)
} else {
throw new Error(`Job mode ${job.mode} not implemented`)
}
}
try {
const result = await vmBackup.run()
task.success(result)
return result
} catch (error) {
vmBackupFailed(error)
}
})
.catch(noop) // errors are handled by logs
}),
error =>
runTask(taskStart, () => {
throw error
getVmTask().run(() => {
vmBackupFailed(error)
})
)
}
const { concurrency } = settings
await asyncMapSettled(vmIds, concurrency === 0 ? handleVm : limitConcurrency(concurrency)(handleVm))
const _handleVm = concurrency === 0 ? handleVm : limitConcurrency(concurrency)(handleVm)
while (queue.size > 0) {
const vmIds = Array.from(queue)
queue.clear()
await asyncMapSettled(vmIds, _handleVm)
}
}
)
}

View File

@@ -12,7 +12,7 @@
- [REST API] Export host [SMT](https://en.wikipedia.org/wiki/Simultaneous_multithreading) status at `/hosts/:id/smt` [Forum#71374](https://xcp-ng.org/forum/post/71374)
- [Home & REST API] `$container` field of an halted VM now points to a host if a VDI is on a local storage [Forum#71769](https://xcp-ng.org/forum/post/71769)
- [Size Input] Ability to select two new units in the dropdown (`TiB`, `PiB`) (PR [#7382](https://github.com/vatesfr/xen-orchestra/pull/7382))
- [Backup] Ability to set a number of retries for VM backup failures [#2139](https://github.com/vatesfr/xen-orchestra/issues/2139) (PR [#7308](https://github.com/vatesfr/xen-orchestra/pull/7308))
### Bug fixes

View File

@@ -27,6 +27,11 @@ const SCHEMA_SETTINGS = {
minimum: 1,
optional: true,
},
nRetriesVmBackupFailures: {
minimum: 0,
optional: true,
type: 'number',
},
preferNbd: {
type: 'boolean',
optional: true,

View File

@@ -141,6 +141,7 @@ const messages = {
removeColor: 'Remove color',
xcpNg: 'XCP-ng',
noFileSelected: 'No file selected',
nRetriesVmBackupFailures: 'Number of retries if VM backup fails',
// ----- Modals -----
alertOk: 'OK',

View File

@@ -189,6 +189,7 @@ const getInitialState = ({ preSelectedVmIds, setHomeVmIdsSelection, suggestedExc
drMode: false,
name: '',
nbdConcurrency: 1,
nRetriesVmBackupFailures: 0,
preferNbd: false,
remotes: [],
schedules: {},
@@ -635,6 +636,11 @@ const New = decorate([
nbdConcurrency,
})
},
setNRetriesVmBackupFailures({ setGlobalSettings }, nRetries) {
setGlobalSettings({
nRetriesVmBackupFailures: nRetries,
})
},
},
computed: {
compressionId: generateId,
@@ -644,6 +650,7 @@ const New = decorate([
inputMaxExportRate: generateId,
inputPreferNbd: generateId,
inputNbdConcurrency: generateId,
inputNRetriesVmBackupFailures: generateId,
inputTimeoutId: generateId,
// In order to keep the user preference, the offline backup is kept in the DB
@@ -756,6 +763,7 @@ const New = decorate([
fullInterval,
maxExportRate,
nbdConcurrency = 1,
nRetriesVmBackupFailures = 0,
offlineBackup,
offlineSnapshot,
preferNbd,
@@ -990,6 +998,17 @@ const New = decorate([
value={concurrency}
/>
</FormGroup>
<FormGroup>
<label htmlFor={state.inputNRetriesVmBackupFailures}>
<strong>{_('nRetriesVmBackupFailures')}</strong>
</label>
<Number
id={state.inputNRetriesVmBackupFailures}
min={0}
onChange={effects.setNRetriesVmBackupFailures}
value={nRetriesVmBackupFailures}
/>
</FormGroup>
<FormGroup>
<label htmlFor={state.inputTimeoutId}>
<strong>{_('timeout')}</strong>

View File

@@ -124,6 +124,8 @@ const NewMirrorBackup = decorate([
setAdvancedSettings({ timeout: timeout !== undefined ? timeout * 3600e3 : undefined }),
setMaxExportRate: ({ setAdvancedSettings }, rate) =>
setAdvancedSettings({ maxExportRate: rate !== undefined ? rate * (1024 * 1024) : undefined }),
setNRetriesVmBackupFailures: ({ setAdvancedSettings }, nRetriesVmBackupFailures) =>
setAdvancedSettings({ nRetriesVmBackupFailures }),
setSourceRemote: (_, obj) => () => ({
sourceRemote: obj === null ? {} : obj.value,
}),
@@ -204,6 +206,7 @@ const NewMirrorBackup = decorate([
inputConcurrencyId: generateId,
inputTimeoutId: generateId,
inputMaxExportRateId: generateId,
inputNRetriesVmBackupFailures: generateId,
isBackupInvalid: state =>
state.isMissingName || state.isMissingBackupMode || state.isMissingSchedules || state.isMissingRetention,
isFull: state => state.mode === 'full',
@@ -231,7 +234,7 @@ const NewMirrorBackup = decorate([
}),
injectState,
({ state, effects, intl: { formatMessage } }) => {
const { concurrency, timeout, maxExportRate } = state.advancedSettings
const { concurrency, timeout, maxExportRate, nRetriesVmBackupFailures = 0 } = state.advancedSettings
return (
<form id={state.formId}>
<Container>
@@ -314,6 +317,17 @@ const NewMirrorBackup = decorate([
value={concurrency}
/>
</FormGroup>
<FormGroup>
<label htmlFor={state.inputNRetriesVmBackupFailures}>
<strong>{_('nRetriesVmBackupFailures')}</strong>
</label>
<Number
id={state.inputNRetriesVmBackupFailures}
min={0}
onChange={effects.setNRetriesVmBackupFailures}
value={nRetriesVmBackupFailures}
/>
</FormGroup>
<FormGroup>
<label htmlFor={state.inputTimeoutId}>
<strong>{_('timeout')}</strong>

View File

@@ -319,6 +319,7 @@ class JobsTable extends React.Component {
compression,
concurrency,
fullInterval,
nRetriesVmBackupFailures,
offlineBackup,
offlineSnapshot,
proxyId,
@@ -349,6 +350,9 @@ class JobsTable extends React.Component {
{compression !== undefined && (
<Li>{_.keyValue(_('compression'), compression === 'native' ? 'GZIP' : compression)}</Li>
)}
{nRetriesVmBackupFailures > 0 && (
<Li>{_.keyValue(_('nRetriesVmBackupFailures'), nRetriesVmBackupFailures)}</Li>
)}
</Ul>
)
},